1

Paper to the Annual Conference of the UK Political Studies Association

Cardiff Wales

25-27 March, 2013

Understanding the Dynamics of Internet-based Collective Action using Big Data:

Analysing the Growth Rates of Internet-based Petitions

Scott Hale, Helen Margetts, Taha Yasseri

Oxford Internet Institute, University of Oxford

{scott.hale,helen.margetts, taha.yasseri}@oii.ox.ac.uk

Abstract

Now that so much of collective action takes place on Internet-based platforms, it leaves a digital

imprint which may be harvested to further understanding of the dynamics of mobilization. This 'big

data' offers social science researchers the potential for new forms of analysis, using real-time

transactional data based on entire populations, rather than sample-based surveys of what people

think they did or might do. This paper uses a big data approach to track the growth of over 30,000

petitions to the UK Government on two platforms over two years, analysing the rate of growth and

testing the hypothesis that the distribution of daily change will be leptokurtic (rather than normal)

as previous research on agenda setting would suggest. This hypothesis is confirmed, suggesting that

Internet-based mobilization is characterized by tipping points (or punctuations) and explaining some

of the volatility in online collective action. We find that the vast majority of petitions fail to attain

even minimal levels of success. Where there is success, the punctuations occur very early in the life of

a petition. We propose a model of ‘collective attention’ to characterise this rapid growth and early

saturation. These findings have implications for the strategies of those initiating petitions and the

design of web sites with the aim of maximising citizen engagement with policy issues.

Keywords Collective Action, Big Data, Petitions, Attention

2

Introduction

Now that so much of collective action takes place on Internet-based platforms, most

mobilizations include a digital element and some take place almost wholly online. All

this Internet-based activity leaves a digital footprint, which allows researchers to generate

‘big data’, real-time transactional data of political behaviour. This kind of data presents

new opportunities for social science research, allowing a move away from sample-based

survey data about what people think they did or will do, to transactional data about what

people actually did based on whole populations. Big data is receiving massive attention

and interest across the corporate world and scientific research communities (Mayer

Schoenberger and Cukier, 2013). Yet the potential of this kind of data for understanding

political behaviour remains under-explored. We are still at the start of exploiting the great

potential of ‘big data’ to explore the dynamics of Internet-based collective action and how

it may challenge existing theories and models of collective behaviour.

This paper analyses two ‘big data’ sets of Internet-based mobilizations, generated

from two electronic petition platforms; first, the e-petition part of the No.10 Downing

Street web site (from February 2009 to the point at which it closed in March 2011) and

second, the new e-petition site developed by the UK Cabinet Office for the incoming

Coalition Government in 2010 and launched in August 2011, which replaced it. E-

petitions were first launched in the UK in November 2006, and over the course of its

lifetime the No. 10 site received more than 8 million signatures from over 5 million

unique email addresses.1 Both sites have allowed anyone to view petitions, and any user

with a valid email address could create a new petition or sign an existing petition. There

are important differences between the sites however; for example, whereas the first site

showed the names of the 500 most recent petitioners, the new site shows only the name

of the petition creator. The sites have also provided alternative measures of ‘success’ of a

petition; for example, for the earlier site, the government promised an official response to

all petitions receiving at least 500 signatures, while the incoming Coalition government in

2010 promised that any petition on the new site attracting more than 100,000 signatures

would qualify for a parliamentary debate on the issue raised.

This paper analyses the growth of petitions and the distinctive characteristics of the

mobilization curves of successful and unsuccessful petitions on both platforms. First, we

provide some background on online collective action in general and e-petitions in

1 http://www.mysociety.org/projects/no10-petitions-website/

3

particular. We use previous work on political attention to develop a hypothesis regarding

the development of online mobilizations: that they will be characterised by long periods

of stasis and short periods of rapid change, leading to a leptokurtic distribution of daily

change. Second, we outline the methods used to test this hypothesis and third, provide

the results for two big data sets, harvested from two e-petition platforms provided by the

UK government, followed by a discussion of the policy and design implications of the

findings.

Background

The 21st century has seen a prominent role for the Internet in mobilization, from the

dramatic events in authoritarian states of the Arab Spring, to a series of protests,

demonstrations and social backlash against austerity driven cutbacks and state

retrenchment in liberal democracies facing the consequences of the financial crash of

2008. Researchers have turned their attention to the theoretical and conceptual

implications of online collective action (e.g. Bennett and Segerberg, 2012; Bimber, 2003;

Lupia and Sin, 2005) and some are using innovative methods, including experiments and

data-mining, to explore the viral spread of mobilizations across online social networks

(see, for example, Ackland, 2007; Etling et al, 2009; Hindman, 2008; Gonzalez Bailon et

al, 2011; Segerman and Bennett, 2011; Aral and Walker, 2010).

While online activity may be a minor element of some mobilizations, other

mobilizations occur almost entirely online. The trend for ‘e-petitioning’ represents one

such activity, where online petitions are created, disseminated, circulated, and presented

online, and although policy-makers may discuss responses in offline contexts, such

responses are generated and sent online. The UK government’s e-petition site was

created by the social enterprise MySociety on the No. 10 Downing Street website in

November 2006 and ran until March 2011, when it was closed by the incoming Coalition

government. Some of these petitions had high policy impact, notably one against the

Labour administration’s proposed road pricing policy, which policy-makers admitted off

the record played a role in getting the policy scrapped. A new site was launched in

August 2011 by the Cabinet Office, initially on the direct.gov portal, which transmuted

into the new portal www.gov.uk in the autumn of 2012, with a different format. Signing

petitions has long been among the more popular political activities, leading the field for

participatory acts outside voting and with other social benefits ascribed to it as well as

4

having the potential to bring about policy change; e-petitioning reinforces ‘civic

mindedness’ (Whyte et al. 2005) and is one of a growing portfolio of Internet-based

democratic innovations (Smith, 2009). The widespread use of e-petitions by both

governments and NGOs such as Avaaz and 38 Degrees has received accolades for their

democratic contribution (Escher, 2011; Chadwick, 2012) and the German e-petition

platforms have been analysed extensively (see Lindner and Riehm, 2011; Jungherr and

Jurgens, 2010), but the UK petition platforms have received rather less attention in recent

political science research, with the exception of Wright (2012).

E-petitions are interesting examples of mobilizations with a strong online imprint,

which will include the entire transaction history for both successful and unsuccessful

mobilizations. The data that can be harvested from the signing of electronic petitions is

an example of what is now commonly known as ‘big data’, representing a transactional

audit trail of what people actually did (as opposed to what people think they did) and an

entire population (without the need to take a representative sample). Data like this

represents a big shift for social science research into political behaviour, which has

traditionally rested on survey data, or, for elections, voting data. Big data also presents

challenges to social science research—it doesn’t come with handy demographics attached

and we do not know where people came from to any one interaction, nor where they are

going, so it is often difficult to match up online activities across different platforms, or to

identify the underlying factors influencing behaviour, such as age, income or gender.

This data however, makes it possible to look at the different patterns of growth in the

30,000 mobilization curves that we have and identify the distinctive characteristic of

those mobilizations that succeed and those that fail with our digital hindsight. Such an

analysis, using data that has rarely been available to political science researchers before

the current decade, may tell us something about the nature of collective action itself in a

digital world. Of the research noted above, Jungherr and Jurgens used a smaller dataset

to illustrate the viability of a big data (or computational social science) approach, but

other studies used surveys (Lindner and Riehm, 2011) or more qualitative approaches

(Wright, 2012).

So what would we hypothesise about these mobilizations? A possible hypothesis may

be derived from previous research on agenda setting in political systems. The most well

known model of how policy attention proceeds in a liberal democracy is that of

‘punctuated equilibrium’, developed by the US authors Baumgartner and Jones and their

‘Policy Agendas’ programme of research (see www.policyagendas.org). The theory

5

argues that policy attention to any issue will remain in long periods of stasis where little

change occurs. Where issues do hit the policy agenda, it will be because some event has

‘punctuated’ the equilibria, all eyes (including the media, public opinion, interest groups

and politicians concerned) turn to the issue, money is spent, institutions are created and

policy change occurs (John and Margetts, 2003; Baumgartner and Jones, 1993; Jones and

Baumgartner, 2005). The theory of punctuated equilibria is multi-faceted and has been

illustrated by a range of empirical data across policy areas and within different

dimensions of attention, such as public opinion, budgetary change and congressional

attention (Baumgartner and Jones, 2005) and in various countries, including the UK

(John and Margetts, 2003). Baumgartner and Jones do not discuss Internet-related

activity to any great degree; however, we might hypothesise that the pattern of

mobilizations around a petition would proceed in a similar way, thereby contributing to

the same sort of issue attention cycle that has been observed many times over in agenda

setting research. Such a model would predict that the distribution of daily changes in

attention would be ‘leptokurtic’, with a small number of large changes and a longer tail of

much smaller ones.

Such a finding could not show any causal effect, as only the activity of petitioners is

being analysed here (and tipping points could suggest a media effect, although it is

extremely unlikely that this could take place before substantive amounts of signatures

had amassed). It could, however, point towards a role for online mobilization in policy

change analogous to that of the media in the agenda setting analysis, which is ascribed a

lurching effect, due to the capacity of the media to parallel process only a small number

of issues; at the point at which a punctuation occurs, media attention will ‘tip over’ from

specialist outlets into the mainstream media. In addition, in lower level mobilizations

where the media is not paying attention, such a finding could suggest a role in policy

issues below the media radar.

Methods

The UK Government's petition website (petitions.number10.gov.uk) was accessed daily

from 2 February 2009 until March 2011, when the site closed, with an automated script.

Each day, the number of overall signatures to date on each active petition was recorded.

In addition, the name of the petition, the text of the petition, the launch date of the

petition, and the category of the petition were recorded. Overall, 8,326 unique petitions

6

were tracked from the earlier site, representing all publically available petitions active at

any point during the study. A first pass at this data after the site closed revealed the

importance of the first day in the future of a petition (see below and Hale and Margetts,

2012) and suggested that more frequent scraping of the data could deliver a more fine-

grained analysis. For this reason, when the new site was launched in August 2011, we set

the automatic script to scrape it every hour, recording the same details as for the previous

site. Our second dataset currently contains hourly data points for all the petitions

(19,789) submitted to the new site from 5th August 2011 to 22nd February 2013.

The two sites differed under different policies for how the government would

respond to petitions. For the No 10 Downing Street site, prospective petitioners were told

that if their petition achieved 500 signatures, they would receive an official response.

There were no other official measures of success, although one petition did succeed in

raising over one million signatures, which previous research has identified as a possible

‘tipping point’ for mobilizations; if potential participants know that more than one

million have participated, they are more likely to participate themselves (Margetts et al,

2011); the petition was widely regarded as influential in getting the policy reversed. For

the Cabinet Office site, the bar for an official response is unclear from the site, although

the majority of petitions that have over 10,000 signatures do receive a response with the

prefix ‘As this e-petition has received more than 10,000 signatures, the relevant

Government department have provided the following response’. More importantly, in the

early days of the Coalition Administration, David Cameron promised that signatures

obtaining more than 100,000 signatures would generate a parliamentary debate on the

issue raised by the petition. All these information cues will have acted as possible drivers

on individuals considering whether to sign a petition.

Petitions on the Number 10 website closed, by default, 12 months after they first

launched. To identify patterns in how petitions grow, the percentage change in new

signatures was calculated each day, for the earlier site. Most petitions had a long period

of inactivity prior to their deadline date. To consider just how petitions grow, data was

truncated after the last signature on a petition, removing any final period of zero

signature-per-day growth prior to the petition's deadline.

Leptokurtic distributions have a more acute peak close to the mean and larger tails.

There is no statistical test to specifically classify a distribution as leptokurtic. However,

several tests in combination help demonstrate a distribution is leptokurtic (see John and

Margetts, 2003). The most rigorous test is the Shapiro-Wilk test (1965), which checks

7

whether the points could possibly be drawn randomly from a normal distribution.

Leptokurtic distributions should reject the Shapiro-Wilk null hypothesis of normality.

The Kolmogorov-Smirnov test (Chakravarti et al. 1967, pp. 392-4) tests that a set of

frequencies is normal distributed by focusing on the skewedness and kurtosis of a

distribution, and this null hypothesis should be rejected if a distribution is leptokurtic and

hence non-normal. Visualizing the histogram and plotting a log-lot plot, which should be

nearly a straight line if changes are leptokurtic, give further evidence of a leptokurtic

distribution.

Results

First, we explored the data harvested from the first e-petitions site on the No. 10

Downing Street web site, which produced a set of 8,326 unique petitions, shown in

Figure 1. The most immediate finding of interest was that 94 per cent failed to obtain

even the modest 500 signatures required to elicit an official response, the only measurable

‘success’ indicator for the earlier site. Nearly all petitions that succeeded in obtaining 500

signatures did so quickly. Successful petitions took a mean time of 8.4 days to reach 500

signatures, but a median time of only two days. In fact, 230 of the 533 successful

petitions succeeded in obtaining 500 signatures on the day they were launched (day 1).

Only a few petitions take a much longer time to reach the 500 signature mark: 31

petitions (6 per cent) succeed after taking more than 30 days, and only five petitions in

our dataset reached the 500 signature mark after being active more than four months.

Next, we tested our hypothesis that the distribution of daily change in signatures

would be leptokurtic. Figure 2 shows the percentage change in new signatures adjusted

so that the mean growth of each petition lies at zero. While most daily change is small,

petitions’ growth is punctuated by a few large changes. The distribution of growth is

leptokurtic and strongly rejects the Shapiro-Wilk null hypothesis of normality with a w

statistic of 0.17 translating to a p-value less than 0.000001. The distribution has a kurtosis

score of 1,445 and a skewedness of 30.53, and rejects the Kolmogorov-Smirnov test for a

normal distribution (p<0.0001). When we applied the same tests to the population of

petitions that were successful in achieving 500 signatures (that is, excluding the

unsuccessful ones), we found a similar leptokurtic distribution (Shapiro-Wilk w statistic

of 0.10, p<0. 000001).

So having identified punctuations, what can we say about where they are? The

8

largest daily changes happened at the start in the life cycle of the petition. Looking at the

distribution of the day on which the punctuation occurred, we see that (for example) all

daily changes of more than 80 per cent occurred in the first five days, and greater than 40

per cent in the first 8 days 0 and even for all changes over 40 per cent, the median day is

1, and the mean is 2.2 and the third quartile is one.

So it seems that the early days of a petition are crucial, in particular the first day.

Running a logit regression revealed that the number of signatures a petition received on

its first day is the most important factor in explaining the petition’s success, and a linear

regression (shown in Table 1) showed that it was also the most important factor in

explaining the total number of signatures the petition receives during its lifetime. All

other factors tested—the topic category, the start day of the week, weekend vs weekday

launch—did not have a significant effect on the growth of a petition once controlling for

the number of signatures the petition received on its first day. Petitions tended to grow

shortly after launch and then stop growing. This active period of growth for petitions has

a mean length of 57 days and a median length of 27 days.

Figure 1. Petition Growth on No. 10 Downing Street e-petitions site, 2009 – 2010

Note: N = 8,326 petitions, all petitions created between September 2009 and May 2010

9

Table 1: Factors affecting growth. OLS predicting the total number of signatures

Estimate Std. Error t value Pr(>|t|)

(Intercept) 218.41 78.55 2.78 0.00544 **

Category -6.37 4.97 -1.28 0.19976

Day of week

petition started

-12.23 11.79 -1.04 0.29974

Signatures

collected on first

day

1.95 0.02 96.29 <0.00001

***

Number of

petitions started

on the same day

-1.55 2.96 -0.53 0.59960

Note: Adjusted R-squared: 0.5268

Figure 2: Log of daily percentage change in number of signatories (centred around

each petition's mean).

10

Note: Any final period in which petitions that gain no more signatures after a certain point before

closing date has been removed from the daily per cent change data (so the tallest bars do not include

these ‘zero change’ days.

The identification of the first day as a key ‘tipping point’ for successful petitions caused

us to re-examine our data collection techniques as we took the data harvesting and

analysis forward with the new site that opened in the summer of 2011. That is, the

mobilization on this first day was so rapid that daily scraping of the site could not

provide a fine-grained analysis of this or other potential tipping points. Therefore, as

noted above, when we collected data from the new site we did so every hour, providing

us with 19,789 petitions up to February 2013 (we continue to collect the data). The

analysis of this dataset in this draft paper, however, uses a randomly selected subset of

this new data. For this subset, Figure 3 shows (in different colours) those petitions that

attained the first level of ‘success’ (that is, the 10,000 signatures required for an official

response); and those that attained the second level, the 100,000 signatures required to

generate a parliamentary debate.

Figure 3: Petition growth on Cabinet Office e-petitions site, 2011 - 2013

Note: Graph shows a fraction of N=3813 petitions, all created between 5th August 2011 and 22nd

February 2013. Note also that y-axis uses a logarithmic scale.

Once again, it is immediately clear that the vast majority of petitions did not achieve any

11

measure of success. Only 5 per cent of petitions obtained the 500 signatures, which we

calculated to compare with the previous dataset, and only 4 per cent received 1,000.

Only 0.7 per cent attained the 10,000 signatures which seems to be the bar for receiving

some sort of official response, and only 0.1 per cent attained the 100,000 required for a

parliamentary debate. Again, the first day was crucial to achieving any kind of success.

Any petition receiving 100,000 signatures after three months, needed to have obtained

3,000 within the first 10 hours on average.

Discussion

The results from both data sets show just how few petitions actually attain success by any

measure. For the earlier data from the No.10 Downing Street site, the 500 signature mark

seems at first consideration a very low threshold that should easily be passed. However,

by far the majority of petitions (94 per cent in this time period) fail to attain even this

modest number of signatures, illustrating the point that in online environments, the low

costs of initiating a collective action mean that there are likely to be large numbers of

unsuccessful mobilizations. Petitions are most active when they are first launched and

most petitions (presumably in the lack of outside stimulus) become digital dust after a

couple of months despite typical deadlines of one year on the site. Our second data set

tells the same story, suggesting that this finding may be generalized to other

mobilizations rather than representing some characteristics of the No. 10 Downing Street

platform. The finding that most mobilizations of this kind fail to take off in any sense

chimes well with recent research into the spread or diffusion of initiatives across online

networks. Goel et al (2012), for example analyse the diffusion patterns arising from

online domains, ranging form networked games to microblogging services and find that

in all their seven cases, the vast majority of cascades are small, and are described by a

handful of simple tree structures that terminate within one degree of an initial adopting

‘seed’. Even for the few large cascades that they observed, the bulk of adoptions often

take place within one degree of a few dominant individuals. Although we have not yet

made an attempt to model the network activity behind the petitions studied here, it seems

likely that to do so would reveal a similar pattern.

We attempt to capture the characteristic of early rapid growth and decay that the data

reveals, with a model of ‘collective attention’ decay, drawing on Wu and Huberman

(2007). In their model, they calculate a ‘novelty’ parameter relating to the novelty of

news items on a news sharing platform, but in a more general framework, the decay in

12

attention could have other reasons, for example reaching the system size limits, or lack of

viral spread. In the model, N agents at the time t, bring Nµ new agents in the next step in

average, µ being a multiplication factor; in our case, this would mean that every signature

brings µ new signatures in the next hour, leading to an exponential growth of rate µ in

the number of signatures. This model works quite well at the beginning, but very soon

the spread rate decays and new signatures come at a much lower rate. Therefore we let

the multiplication factor decay by introducing a second factor r(t), which decays in a way

that is an intrinsic of the medium; each signature at time t, on average brings µr(t) new

signatures in the next hour. To correct for the early saturation observed in the empirical

results, we enter an ‘outreach’ parameter which can change over time and damp the fast

initial growth. To calculate that empirically, we average over the logarithm of the

number of signatures in hourly bins, starting from the launching time and then calculate

the hourly increment at time t and normalize it by the logarithm of the number of

signatures up to time t as follows:

In other words, the outreach factor measures the relative growth of the logarithm of

number of signatures within an hour, averaged over the whole sample. This parameter is

shown as a function of adjusted time in Figure 4. This shows that collective attention

decays very fast indeed, and that after 24 hours, a petition’s fate is virtually set.

13

Figure 4. Rate of Change of ‘Collective Attention’ paid to Petitions created on the

Cabinet Office site calculated based on Eq. 1.

Confirmation of our hypothesis regarding the leptokurtic distribution of changes to

the support for a petition suggests that in online environments, collective action could

play a role in a punctuated equilibrium model of policy change. That is, the general

pattern for policy attention is for issues to remain dormant or in stasis, with a generally

low level of attention. Some issues (by far the minority) that attract attention quickly gain

a ‘critical mass’ of activists and start to vie for policy attention, joining the range of other

institutional influences in helping to ‘punctuate’ the equilibrium. Such an argument

would not include the claim that the mechanism by which collective action acts to bring

about instability would be the same as the role played by the media, which plays a

distinctive ‘lurching’ role in Jones and Baumgartner’s analysis, based in part on the

tendency of the media to process a small number of issues in parallel. In the context

studied here, the mechanism would depend more on the ways in which a mobilization is

disseminated via online social networks, something that previous research mentioned in

the introduction has begun to investigate. If such activity tends to take place on the day

14

the petition is initiated, then these findings could indicate the importance of achieving

some kind of viral spread across the petitioner’s closest contacts right at the start, because

initial rapid growth will have a greater effects on subsequent participation decisions by

‘weaker tie’ contacts than a gradual growth over a prolonged period.

In other empirical studies using experimental methodologies, we have started to

uncover the mechanism behind such punctuations. That is, in addition to the close

contacts informed about the petition directly by the petitioner, other individuals

(including people unknown to the petitioner or more distant contacts) deciding whether

to participate will be influenced by the information that other people have already

participated. This influence will depend on the number of other participants (Margetts et

al, 2011), the personality of the individual deciding whether to participate (Margetts et al,

2013), and the closeness of the individual to the petitioner. The occurrence of a

punctuation will depend on the existence of ‘starters’ whose thresholds for participation

are low or whose closeness to the petitioner has in this instance reduced their threshold

for participation. These starters will act as a signal for people with higher thresholds and

weaker ties to the petitioner to ‘follow’ in signing the petition, thereby acting as a further

signal for people with even higher thresholds to join. At some point, if the petition is

successful, then the number of followers will reach ‘critical mass’ and attention to the

mobilization will become widespread, breaking out of the petitioner’s social network and

gain more general social media exposure.

So, there are two possible explanations for the importance of the first day in

achieving the critical mass: first, the way in which the petition is disseminated via online

networks and second, the dynamic of starters and followers and social information about

other participants in the burst of activity. If it is that visitors to the petition are heavily

influenced by the numbers of other participants that they observe, then the finding could

join those of other work in economics illustrating the importance of ‘first donations’ in

charitable giving (Bog et al, 2006), showing that early donors set the precedent for later

donors, or the wider literature on conditional co-operation, showing that social

information about the contribution of others influences an individual participant’s

decision to contribute (Croson and Shang, 2009; Frey and Meier, 2004). This study

complements this previous work by focusing on numbers of participants, rather than

contribution amount (as everyone’s contribution, at least that we are able to measure, is

the same) and a more explicitly political context, rather than that of charitable giving.

The strong effect of petitions tending to succeed quickly or not at all will be

15

influenced by the design of the petition website during the period of study; the ‘outreach’

factor will vary across platforms. For users starting at the homepage of the earlier No.10

Downing St site, it was possible to view petitions overall or within a specific category and

to sort petitions by the number of signatures or the date added. It was therefore easiest to

look at petitions with the largest or smallest number of signatures and the oldest or

newest petitions. On the newer Cabinet Office site, petitions can be sorted by signature or

closing date, or viewed by government department, but not by topic. In addition, we can

expect different behaviour from users of either site who arrive at the homepage (who may

respond to these information cues by looking only at the newest petitions or the petitions

with the most signatures contributing to the effects observed), or users following links

shared via email and social media, which would point to a specific petition that the

contact was supporting (who will avoid these information cues). These alternative effects

might be tested using an experimental approach in future research. In March 2012, the

Cabinet Office introduced a change to the e-petitions site which also shaped the

information environment of prospective petitioners, by introducing a ‘Trending e-

petitions’ with the highest signatures on the front page. The fact that our data was

captured both before and after this change provides us with a ‘natural experiment’

whereby we can test the effect of this change, which will be the next step in this

programme of research.

Conclusions

We have found that in online mobilizations, growth tends not to occur, meaning that

most mobilizations that are initiated, fail. But where it does, it proceeds in rapid bursts

followed by periods of stasis. Such a finding suggests that online mobilizations of the

kind covered here could play a role in the more general process of punctuated equilibria

in policy-making. For example Jones and Baumgartner (2005) found a high correlation

between public concern on an issue and Congressional attention. Our findings here could

be even more interesting. In the theory of punctuated equilibrium the media plays a key

role in terms of ‘lurching’ from one issue to another and having a complex feedback

relationship with public opinion. But the sort of mobilizations we are looking at here are

bubbling up relatively independently of the media, gaining media attention only when

they obtain significantly high levels of support—the petition on road-pricing that

successfully played a role in obtaining policy change, for example, received a great deal

16

of media attention once that it reached one million signatories. Research that develops

our understanding of the mechanics of this turbulence will be important for scholars and

policy-makers alike as collective action continues to move into online settings. If online

collective action is characterised by punctuations, then it looks as if such activity could

inject a further dose of instability into political systems. If mobilizations follow a pattern

of very low levels of attention punctuated by occasional ‘spurts’ which grow rapidly into

full scale mobilizations that merge with other elements of the political system to push

policy change on to the agenda and the institutional landscape, then we can expect to see

increasing turbulence in contemporary politics, adding to the ‘instability’ that

Baumgartner and Jones (1993; 2005) and their co-investigators have modelled so

extensively in previous research.

This paper has, we hope, demonstrated the potential for ‘big data’ approaches in

political science research. The data we report here was automatically and non-obtrusively

generated to provide a dataset of real-time transactional data of a kind that has rarely

been available to political science researchers before. One of the aims of the research

programme of which this analysis forms a part is to develop the methods we have used to

both harvest and analyse the data, which require skills and expertise and conceptual

approaches that span academic disciplines; of the authors of this paper, one is a physicist,

one a computer scientist and one a political scientist. As big data is used more extensively

in this kind of research, the ability to work across disciplines in this way will become

increasingly important.

Our future empirical work will also explore co-ordination with media coverage and

mentions of the petitions on social networking sites (such as Facebook, YouTube,

Google Search and Twitter); we are now systematically gathering data on any mention of

the petitions for which we have captured data across all these platforms. Much work on

online collective action is based on research into a single platform, whereas any online

activity tends to involve several. By looking carefully at the timing with which an issue

gains attention in different parts of the political system, including the activist activities

investigated here, we might get closer to establishing some sort of sequencing of

attention. In addition, one way of getting around the lack of causal inference in research

of this kind is to carry out experiments, as in Margetts et al (2011, 2012, 2013). Future

work will also use an experimental approach to analyse the effect of different information

environments surrounding petition websites on petition growth. Designers of web sites

that involve civic engagement, such as e-petition sites, must decide what social

17

information about existing levels of participation to include, for example the numbers of

people who have already signed and the timings of when they did so. Other design

decisions include whether participants are anonymous (as on the Cabinet Office site) or

whether their names are made visible (as on the earlier No. 10 Downing St site and the

German e-petitions platform), and whether input from other social media platforms is

incorporated into the petition site. Research of this kind, using both ‘big data’ and

experimental approaches, can inform such design decisions in ways that maximise

citizens’ input to policy debates.

References

Ackland, R. And Gibson, R. (2006) Hyperlinks and horizontal political communication on the

WWW: The untold story of parties online. The Australian National University.

Aral, S. and Walker, D. (2011) ‘Creating Social Contagion Through Viral Product

Design: A Randomized Trial of Peer Influence in Networks, Management Science.

Baumgartner, F. and Jones, B. (1993) Agendas and Instability in American Politics.

University of Chicago Press pp 1-298.

Bennett, L. and Segerberg, A. (2011) ‘Digital Media and the Personalization of

Collective Action: Social Technology and the Organization of Protests against the

Global Economic Crisis.’ Information, Communication & Society (2011) 14: 770-799.

Bennett, L., and Segerberg, A. (2012) ‘The Logic of Connective Action’, Information,

Communication & Society, 15 (5): 739-768.

Bimber, B. (2003) Information and American democracy: Technology in the evolution of political

power. New York: Cambridge University Press.

Bimber, B, Flanagin, A, Stohl, C (2006) ‘Reconceptualizing Collective Action in the

Contemporary Media Environment’, Communication Theory, Vol. 15 Iss. 4: 365-388.

Bøg, M., Harmgart, H., Huck, S. and Jeffers, A. M. (2012), Fundraising on the Internet.

Kyklos, 65: 18–30. doi: 10.1111/j.1467-6435.2011.00525.x

Borge, R. and Cardenal, A. (2011) ‘Surfing the Net: A pathway to participation for the

politically uninterested?’, Policy and Internet, vol 3, issue 1.

Centola, D. (2010) ‘The Spread of Behaviour in an Online Social Network Experiment’

Science 329, 1194.

Croson, Rachel and Jennifer. Shang (2009). ‘Field Experiments in Charitable

18

Contribution: The Impact of Social Influence on the Voluntary Provision of Public

Goods’, Economic Journal. 119 (540): 1422–1439.

Escher, T. 2011. Writetothem.com: Analysis of users and usage for UK Citizens Online

Democracy, London: My Society.

Etling, B. Kelly, J. Faris, R. and Palfrey, J. (2009) Mapping the Arabic Blogosphere: Politics,

Culture and Dissent, Berkman Center Research Publication NO. 2009-06.

Frey, Bruno S. and Stephan, Meier (2004). Social Comparisons and Pro-social Behavior:

Testing Conditional Cooperation in a Field Experiment, American Economic Review.

94: 1717–1722.

Goel, S. Watts, D. and Goldstein, D. (2012) ‘The Structure of Online Diffusion

Networks’, Proceedings of the 13th ACM Conference on Electronic Commerce, pp. 623-638.

Gonzalez Bailon, S. Borge-Holthoefer, J. Rivero, A. and Moreno, Y. (2011) ‘The

Dynamics of Protest Recruitment through an Online Network’, Scientific Reports,

December.

John, P., & Margetts, H. (2003). Policy punctuations in the UK: Fluctuations and

equilibria in central government expenditure since 1951. Public Administration, 81(3),

411-432. Blackwell Publishing Ltd. doi:10.1111/1467-9299.00354

Jones, B. D., & Baumgartner, F. R. (2004). Representation and Agenda Setting. Policy

Studies Journal, 32(1), 1-24. Blackwell Publishing Ltd. doi:10.1111/j.0190-

292X.2004.00050.x

Jones, B. and Baumgartner, F. (2005) The Politics of Attention: How Government Prioritizes

Problems, Chicago: University of Chicago Press.

Jungherr, A. Jurgens, A. (2012) ‘The Political Click: Political Participation through E-

Petitions in Germany’, Policy and Internet: 2 (4). Pp. 131-165.

Liben-Nowell, D., & Kleinberg, J. (2008). Tracing information flow on a global scale

using Internet chain-letter data. Proceedings of the National Academy of Sciences, 105(12),

4633. National Acad Sciences.

Lindner, R. and Riehm, U. (2011) ‘Broadening Participation Through E-Petitions? An

Empirical Study of Petitions to the Germany Parliament’, Policy and Internet: 3(1), pp.

63-85.

Lupia, A., & Sin, G. (2003). Which Public Goods are Endangered?: How Evolving

Communication Technologies Affect The Logic of Collective Action. Public Choice,

117(3), 315-331. Springer Netherlands. Retrieved from

19

http://dx.doi.org/10.1023/B:PUCH.0000003735.07840.c7

Margetts, H., John, P., Reissfelder, S., and Hale, S. 'Applying Social Influence to

Collective Action: Heterogeneous Personality Effects' (July 19, 2011,

http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1892805)

Margetts, H. John, P., Escher, T. and Reissfelder, S. (2011) 'Social Information and

Political Participation on the Internet: an Experiment', European Political Science

Review.

Margetts, H., John, P., Escher, T., & Reissfelder, S. (2009) Experiments for Web

Science: Examining the Effect of the Internet on Collective Action, Proceedings of

WebSci2009 Conference, Athens, 18-20 March.

Mayer-Schoenberger, V. and Cukier, K. (2013) Big Data: A Revolution That Will Transform

How We Live, Work and Think. John Murray.

Mosca, L., & Santucci, D. (2009). Petitioning Online. The Role of E-Petitions in Web

Campaigning. Political Campaigning on the Web, 121. transcript Verlag.

Schelling, T. C. (2006). Micromotives and macrobehavior. WW Norton & Company.

Smith, G. (2009). Democratic innovations: Designing institutions for citizen participation.

Cambridge, England: Cambridge University Press

Whyte, A., Renton, A., & Macintosh, A. (2005). e-Petitioning in Kingston and Bristol

Evaluation of e-Petitioning in the Local e-Democracy National Project.

Wright, S. (2012) Assessing (e-) Democratic Innovations: “Democratic Goods” and

Downing Street E-Petitions”, Journal of Information Technology and Politics, 9: 453-470.

Wu, F. and Huberman, B. (2007) ‘Novelty and collective attention’, Proceedings of the

National Academy of Sciences of the USA, 104 (45) 17599-17601.